A'oa'oga TensorFlow mo Tagata Amata
O le TensorFlow o se tasi o faʻavae sili ona lauiloa mo le aʻoaʻoina loloto ma le aʻoaʻoina masini. Na atiaʻe e le 'au a le Google Brain, o le TensorFlow ua faʻaaogaina lautele i le tele o poloketi suʻesuʻega ma faʻaoga faʻapisinisi. O lenei tusiga o loʻo tuʻuina atu ai se aʻoaʻoga laʻasaga i lea laʻasaga e fesoasoani ia te oe, i le avea ai ma se tagata amata, e amata ai i le TensorFlow.
1. Malamalama i Mataupu Faavae o le TensorFlow
A'o le'i amataina ona fa'apipi'i ma fa'aaoga le TensorFlow, e taua le malamalama i le uiga o le TensorFlow ma ona manatu fa'avae. O le TensorFlow o se fa'avae tatala mo le fa'atusatusaga fa'afuainumera ma le a'oa'oina o masini. E fa'aaogaina kalafi tafe o fa'amaumauga e fa'atino ai galuega fa'afuainumera, lea e fai ma sui o nodes i le kalafi galuega fa'amatematika, ma o pito e fai ma sui o fa'asologa fa'amaumauga e tele itu (tensors) e feso'ota'i i le va o latou.
2. Fa'apipi'iina o le TensorFlow
O le laasaga muamua i le faʻaaogaina o le TensorFlow o le faʻapipiʻiina lea. O le auala lea e faʻapipiʻi ai le TensorFlow e faʻaaoga ai le pip, le Python package manager.
1. Fa'apipi'iina o le Python:
Ia mautinoa ua fa'apipi'i le Python i lau masini komepiuta. E fetaui lelei le TensorFlow ma le Python 3.6 e o'o atu i le 3.9 i le taimi o lenei tusitusiga. E mafai ona e siiina mai le Python mai le upega tafa'ilagi aloaia a le Python.
2. Siosiomaga Fa'apitoa:
E matuā fautuaina lava le fatuina o se siosiomaga fa'apitoa e vavae'ese ai lau galuega faatino TensorFlow:
"sh"
python -m venv myenv
puna myenv/bin/activate Mo tagata faʻaoga Mac/Linux
myenv\Scripts\activate Mo tagata fa'aoga Windows
""
3. Fa'apipi'iina o le TensorFlow:
Ia, faʻapipiʻi le TensorFlow e faʻaaoga ai le pip:
"sh"
pip fa'apipi'i tensorflow
""
3. Talofa Lalolagi faatasi ai ma le TensorFlow
O lea la ua uma ona fa'apipi'i le TensorFlow, se'i o tatou fatuina se tusitusiga faigofie a le Python e fa'amaonia ai le fa'apipi'iina. Fausia se faila Python fou ma faaigoa i le `hello_tensorflow.py`.
“`python
faaulufale mai tensorflow pei tf
Fausia se mea e tumau
talofa = tf.constant('Talofa, TensorFlow!')
Amata le sauniga
fa'atasi ai ma le tf.Session() e pei o le sess:
iʻuga = sess.tamoʻe(talofa)
lolomi(i'uga)
""
Fetuunai le code e tusa ai ma le TensorFlow version 2.x:
“`python
faaulufale mai tensorflow pei tf
Fausia se mea e tumau
talofa = tf.constant('Talofa, TensorFlow!')
Fa'agaoioi e fa'aaoga ai le eager execution (ua fa'agaoioia e ala i le fa'aletonu)
lolomi(talofa.numpy())
""
Sefe le faila, ona fa'atino lea:
"sh"
python hello_tensorflow.py
""
4. Malamalama i Tensors ma Galuega Fa'avae
O Tensors o le fausaga autū lea o faʻamaumauga i le TensorFlow, o ni faʻasologa e tele itu. O nisi nei o faʻataʻitaʻiga e fesoasoani ia te oe e malamalama ai i tensors:
“`python
faaulufale mai tensorflow pei tf
Fausiaina o tensors
scalar = tf. tumau(7) scalar
vektera = tf. tumau([1, 2, 3]) vektera
matrix = tf.constant([[1, 2], [3, 4]]) matrix
tensor3d = tf.constant([[[1, 2, 3], [4, 5, 6]], [[7, 8, 9], [10, 11, 12]]]) 3D tensor
lolomi(f'Scalar: {scalar}')
lolomi(f'Vector: {vector}')
lolomi(f'Matrix: {matrix}')
lolomi(f'Tensor 3D: {tensor3d}')
""
Mo le faʻatinoina o galuega faʻavae i luga o tensors:
“`python
a = tf.constant([[1, 2], [3, 4]])
b = tf.constant([[5, 6], [7, 8]])
Fa'agaioiga fa'aopoopo
fa'aopoopo = tf.fa'aopoopo(a, b)
Galuega fa'atelevave o le matrix
mul = tf.matmul(a, b)
lolomi(f'Fa'aopoopoga: {fa'aopoopo}')
lolomi(f'Fa'atelega o le Matrix: {mul}')
""
5. Fausiaina o se Faʻataʻitaʻiga Faigofie o le Neural Network
O le isi laasaga o le fatuina lea o se faʻataʻitaʻiga faigofie o fesoʻotaʻiga neura. O le a tatou fausia se faʻataʻitaʻiga faʻavasegaga o ata e faʻaaoga ai le seti o faʻamaumauga MNIST, o se faʻamaumauga o ata numera na tusia i lima. Seʻi o tatou amata:
“`python
faaulufale mai tensorflow pei tf
mai le tensorflow.keras fa'aulufale mai fa'amaumauga, vaega, fa'ata'ita'iga
La'uina mai o le fa'amaumauga a le MNIST
(ata_o_teine, fa'ailoga_o_teine), (ata_o_teine, fa'ailoga_o_teine) = seti_ta'iala.mnist.uta_fa'amaumauga()
Fa'atulagaina o ata
ata_fa'ata'ita'i, ata_fa'ata'ita'i = ata_fa'ata'ita'i / 255.0, ata_fa'ata'ita'i / 255.0
Faia o se faʻataʻitaʻiga
fa'ata'ita'iga = fa'ata'ita'iga.Fa'asologa([
layers.Flatten(input_shape=(28, 28)),
layers.Dense(128, activation='relu'),
lapisi. mafiafia(10)
])
Tuufaatasiga o faʻataʻitaʻiga
model.compile(optimizer='atama',
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
fua fa'atatau=['sa'o'])
Aoaoina o le faʻataʻitaʻiga
model.fit(train_images, train_labels, epochs=5)
Su'eina o le fa'ata'ita'iga
leiloa_o_suega, acc_o_suega = faʻataʻitaʻiga.evaluate(ata_o_suega, faʻailoga_o_suega)
lolomi(f'Sa'o o le suega: {test_acc}')
""
Fa'amatalaga:
– Seti Fa'amaumauga: Matou te fa'aulufale mai ma uta le seti fa'amaumauga MNIST.
– Fa'agasologa Muamua: Fa'atulaga le fa'amaumauga e ala i le vaevaeina o tau o pixel i le 255.
– Faʻataʻitaʻiga: Matou te faʻamatalaina se faʻataʻitaʻiga faigofie e lua vaega. O le vaega muamua o le vaega `Flatten` e liua ai le ata 2D i se faʻasologa 1D. O le vaega lona lua o le vaega `Dense` e 128 neurons ma le `relu` o le galuega faʻagaoioia, ma le vaega mulimuli o le vaega `Dense` e 10 neurons o loʻo fai ma sui o vasega e 10.
– Tuufaatasia: Matou te tuufaatasia le faataitaiga e faaaoga ai le `adam` optimizer ma le `SparseCategoricalCrossentropy` o le galuega faatino o le leiloa.
– A'oa'o: A'oa'o le fa'ata'ita'iga mo le 5 vaitaimi.
– Iloilo: Iloilo le faʻataʻitaʻiga e faʻatusatusa i faʻamaumauga o suʻega.
6. Fa'asaoina ma le Utaina o Fa'ata'ita'iga
A uma ona e toleniina se faʻataʻitaʻiga, atonu e te manaʻo e sefe mo se isi taimi e faʻaaoga ai e aunoa ma le toe toleniina. O le auala lea e sefe ai ma uta ai se faʻataʻitaʻiga:
“`python
Sefeina o le faʻataʻitaʻiga
fa'ata'ita'iga.sefe('la'u_fa'ata'ita'iga.h5')
Fa'ata'ita'iga o lo'o utaina
fa'ata'ita'iga_fou = tf.keras.models.load_model('la'u_fa'ata'ita'iga.h5′)
Fa'amaonia le fa'ata'ita'iga ua utaina
leiloa, acc = new_model.evaluate(test_images, test_labels)
lolomi(f'Sa'o o le fa'ata'ita'iga ua utaina: {acc}')
""
I'uga
O lenei taʻiala o loʻo tuʻuina atu ai se faʻatomuaga auiliili i le amataina o le TensorFlow mo ē amata. Ua matou talanoaina le faʻapipiʻiina, faʻagaioiga faʻavae o le tensor, ma le fausiaina o se faʻataʻitaʻiga faigofie o fesoʻotaʻiga neural e faʻaaoga ai le faʻamaumauga MNIST. E ofoina atu e le TensorFlow le tele o agavaʻa faʻapitoa e suʻesuʻe, e pei o le faʻagasologa faʻapitoa o faʻamaumauga, faʻataʻitaʻiga sili atu ona faigata, ma le faʻaaogaina o le TensorFlow i masini e pei o TPU ma GPU. Matou te faʻamoemoe o lenei aʻoaʻoga e fesoasoani ia te oe e amata ai i le lalolagi o le aʻoaʻoina o masini faʻatasi ai ma le TensorFlow.
